# SAR-ATR-From-Beginning-to-Present: a fifty year survey of radar target recognition > This repository accompanies a 2026 IEEE Geoscience and Remote Sensing Magazine review of fifty years of synthetic aperture radar automatic target recognition, and it collects the papers, datasets, benchmarks, code and pretrained weights that the review points at, updated every two months. It suits a researcher entering SAR ATR or looking for the code behind a method, and it is a curated companion to a paper rather than a library. **JoyeZLearning/SAR-ATR-From-Beginning-to-Present** — Fifty Years of SAR Automatic Target Recognition: The Road Forward （IEEE GRSM 2026） - Repository: https://github.com/JoyeZLearning/SAR-ATR-From-Beginning-to-Present - Stars: 971 · Forks: 3 - Language: Unknown - License: not declared - Published: 2026-10-19 · Updated: 2026-10-19 · Language: en - Canonical page: https://hysenlabs.com/projects/joyezlearning-sar-atr-from-beginning-to-present ## A review paper that needed a repository to point at The paper is Fifty Years of SAR Automatic Target Recognition: The Road Forward, published in IEEE Geoscience and Remote Sensing Magazine in 2026, and the README states plainly that it provides the first comprehensive review of fifty years of synthetic aperture radar automatic target recognition development, tracing its evolution from inception to the present day. The authors are Jie Zhou, Yongxiang Liu, Li Liu, Weijie Li, Taoli Yang, Bowen Peng, Yafei Song, Gangyao Kuang and Xiang Li, and most of them are linked to Google Scholar profiles. The paper is the artefact, and the repository is the companion to it. The README's own framing is a collection of papers, datasets, benchmarks, code and pretrained weights for SAR ATR, released in September 2025, with updates described as currently done bi-monthly, adding interesting and classic articles and datasets, with the update dates and content recorded and others welcome to join in. That last detail is the one to hold on to. This is not a snapshot, it is a growing index, and the update cadence is stated in the README itself. The top level listing is correspondingly thin, holding a README.md and a single directory whose name is 2025_10_update content, which is the accumulated content from the October 2025 update. There is no package, no test suite and no build. The primary language is not determined for the same reason, because there is no program. The readme also asks for stars to encourage more open source work on SAR image interpretation, which is a reasonable request for a project whose whole contribution is curation. ## The gap the review is trying to close is measured in the figures The figures in the README are not decoration, they are the argument. Figure 1 is titled Importance of SAR ATR, and its first panel makes a quantitative claim about the state of the field. From 2020 to 2024, the annual number of published papers in remote sensing and in computer vision reached a similar level, with the gap less than 10 percent, but the number of public remote sensing related code repositories on GitHub remains limited, accounting for only approximately one fourth of the computer vision figure. The stated conclusion is that this discrepancy highlights significant untapped potential for advancing the open source ecosystem within the remote sensing community. All the paper counts come from the WOS Core Collection Database. The other panels of Figure 1 give the most frequent keywords in remote sensing papers over the same period, with the size of each word proportional to its frequency, and the readme names synthetic aperture radar, image classification and object detection as the concepts that have attracted the most attention. The third panel is the application case. SAR ATR is described as widely used and irreplaceable for polar sea ice monitoring and navigation safety in the field of glaciers, for extraterrestrial geology and target recognition in deep space exploration, for forest, flood and deformation monitoring related to global change, and for situational awareness in public safety and national defence. That framing matters for the repository's purpose. The argument for a companion collection is not that the paper needs somewhere to link, it is that a field with computer vision's publication volume and a quarter of its code has a documentation problem, and the review plus a maintained index is the proposed remedy. ## A timeline and an evolutionary tree of the field Figure 2 is a timeline of the evolution of SAR automatic target recognition, and its framing is the history of the field in four phases. It covers the two core tasks of classification and detection, and it runs from understanding physics, through designing features, through learning features, to understanding and learning features at once. That last pairing is the point of the taxonomy, because a method that both understands the physics and learns features is a different animal from one that only does either. Figure 3 turns the timeline into a tree, an evolutionary tree of SAR ATR technology from the 1990s to the present, organised into four primary branches by target type. The branches are ships, vehicles, aircraft and other targets. Branch nodes represent landmark methodologies, connecting lines indicate technological inheritance and innovation, and cross branch linkages signal the emergence of generalisable models. The cross branch linkages are the interesting part for anyone assessing where the field is going. A model that generalises across target types is a different proposition from one tuned to a single class, and a tree that draws those linkages explicitly is making a claim about which works generalised. Figure 4 then defines the task itself, with detection covering the location of potential target regions within a large scale SAR image and classification covering the specific category of a detected target, illustrated with an oil tanker ship, alongside a comparison of optical and SAR images and some challenging instances during SAR target recognition. The two challenges named there are the reason the field exists. Optical imagery has texture and colour that a human or a model can use, while a SAR image is a speckle pattern derived from reflected radar energy, and reading a target from it requires understanding the physics of what the shape does to the returned signal. ## The repository's structure is a table of contents for the field The README's table of contents is the index into the collection, and the two entries visible at the top tell you the shape of it. The first is a summary of surveys in SAR ATR, which is a natural opening for a collection that exists to point at prior art, and the second is SAR target datasets. The dataset section is where an open collection earns its keep in this field. SAR data is awkward to obtain compared with, say, natural image data. A ship detection dataset requires a sensor, a flight line, and access to the resulting image products, and the public ones are usually restricted to specific research groups or available on request. So a curated list of datasets is genuinely useful, and a list that is refreshed every two months is more useful still, because new collections appear and old ones get taken down. The remaining sections named in the table of contents extend the same idea to benchmarks, to code, to pretrained weights and to papers, with the reading of the Chinese text in the README making the intent explicit. The stated aims include exploring how classical knowledge can be inherited, extended and reinvented in the AI era, clarifying which challenges are solved, partly solved, still unresolved or only now emerging, building an open source ecosystem to help push the field forward, and, in the present, looking back to inform the future. That is a list of goals for a community resource, not for a software project, and it describes exactly what a well curated index can do. The citation block is the other structural element. The README asks readers to cite the paper, and gives a BibTeX entry with the journal, the year, the pages and a DOI, so the review is treated as the citable artefact even as the collection continues to grow around it. ## Where this collection is the wrong tool Three cases argue against it. The first is a developer who wants a working system. There is no code to install and no API to call, and the pretrained weights the collection points at are hosted in the upstream projects rather than served from here, so assembling a pipeline means visiting each repository separately and resolving its own dependencies. If your goal is to run SAR ATR on an image today, you want the implementation repository, not the index. The second case is someone who needs a stable, versioned dataset. A collection refreshed bi-monthly is a moving target by design, and the README is candid that update dates and content are recorded, which is a way of saying the list changes. The 2025_10_update content directory shows the shape of the accumulation, and a team that pins a dataset version in a paper or a benchmark will have to do that pinning themselves, against a list that is refreshed on a fixed rhythm. The third case is a reader who wants the survey itself rather than a map of it. The README repeatedly points to the paper for more details, and the paper is where the arguments, the taxonomy and the analysis live, with the repository as the companion. The figures are informative but they are summaries, and the depth is in the IEEE Xplore document. There is also a licensing question that the repository does not answer, and it matters more here than usual. The README states no licence for the repository itself, and the collection is a mixture of things with different terms. A list of links is easier to redistribute than a list of datasets and model weights, and a team building a training pipeline on top of it needs to check the terms of each individual dataset and each set of weights rather than assuming a single licence covers the collection. On the state of the work, the review was accepted by IEEE Geoscience and Remote Sensing Magazine on 2026-08-04, the repository was released in September 2025, and the last push was on 2026-10-05. ## Conclusion SAR-ATR-From-Beginning-to-Present suits a researcher who needs a map of the SAR automatic target recognition field and the assets behind it, rather than one method, and it is especially useful for finding the public code and weights that the field's papers so often omit. It does not suit someone who needs a production inference stack or a maintained library, because the repository is a reading and download list with a README, and it does not suit a user who needs a stable API. Before you plan around it, read the review itself, since the repository's value is a pointer to it, and note that the last push was on 2026-10-05 with the collection described as updated bi-monthly. ## FAQ ### What is the SAR-ATR-From-Beginning-to-Present repository? It is the companion repository for the paper Fifty Years of SAR Automatic Target Recognition: The Road Forward, published in IEEE Geoscience and Remote Sensing Magazine in 2026, and it collects papers, datasets, benchmarks, code and pretrained weights for synthetic aperture radar automatic target recognition. ### How often is the SAR ATR collection updated? The README says the updates are currently done bi-monthly, adding interesting and classic articles and datasets, with the update dates and content recorded and contributions welcome. ### What does the review cover? It traces fifty years of SAR automatic target recognition, with the timeline organised from understanding physics, through designing features and learning features, to understanding and learning features at once, and an evolutionary tree branching by target type into ships, vehicles, aircraft and other targets. ### Why is there so little open source code for SAR recognition? Figure 1 of the paper argues that from 2020 to 2024 remote sensing and computer vision published similar numbers of papers, with the gap under 10 percent, but public remote sensing code repositories on GitHub are only about a quarter of the computer vision figure, which the authors describe as untapped potential for the field's open source ecosystem. ## Sources - [Issues](https://github.com/JoyeZLearning/SAR-ATR-From-Beginning-to-Present/issues) - [JoyeZLearning/SAR-ATR-From-Beginning-to-Present on GitHub](https://github.com/JoyeZLearning/SAR-ATR-From-Beginning-to-Present) - [README](https://github.com/JoyeZLearning/SAR-ATR-From-Beginning-to-Present/blob/main/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/joyezlearning-sar-atr-from-beginning-to-present